Learning Semantic Proxies from Visual Prompts for Parameter-Efficient Fine-Tuning in Deep Metric Learning

Deep Metric Learning (DML) has long attracted the attention of the machine learning community as a key objective. Existing solutions concentrate on fine-tuning the pre-trained models on conventional image datasets. As a result of the success of recent pre-trained models trained from larger-scale datasets, it is challenging to adapt the model to the DML tasks in the local data domain while retaining the previously gained knowledge. In this paper, we investigate parameter-efficient methods for fine-tuning the pre-trained model for DML tasks. In particular, we propose a novel and effective framework based on learning Visual Prompts (VPT) in the pre-trained Vision Transformers (ViT). Based on the conventional proxy-based DML paradigm, we augment the proxy by incorporating the semantic information from the input image and the ViT, in which we optimize the visual prompts for each class. We demonstrate that our new approximations with semantic information are superior to representative capabilities, thereby improving metric learning performance. We conduct extensive experiments to demonstrate that our proposed framework is effective and efficient by evaluating popular DML benchmarks. In particular, we demonstrate that our fine-tuning method achieves comparable or even better performance than recent state-of-the-art full fine-tuning works of DML while tuning only a small percentage of total parameters. Metric learning is a crucial part of machine learning that creates distance functions based on the semantic similarity of the data points. Modern Deep Metric Learning (DML) uses deep neural networks to map data to an embedding space where similar data are closer. This is especially useful for computer vision applications, including image retrieval (Lee et al., 2008; Yang et al., 2018), human re-identification (Wojke & Bewley, 2018; Hermans et al., 2017), and image localization (Lu et al., 2015; Ge et al., 2020). It was recently discovered that the more advanced vision transformers (ViT) are better in terms of representation capability and performance on DML tasks (El-Nouby et al., 2021; Ramzi et al., 2021; Patel et al., 2022; Ermolov et al., 2022).